Optimize LSP operations in nimsuggest for O(1) lookups

[kobi2187]

Faster NimSuggest. Boasting AI description follows, take with grain of salt:

Major LSP performance improvements by replacing O(n) scans with O(1) caching:

1. Symbol location cache for ideUse/ideDus

   • Build hash table mapping symbol ID to all locations
   • Replaces O(n) iteration through ALL symbols across ALL files
   • Cache built lazily on first use
   • Invalidated on recompilation to stay fresh

2. Optimized ideHighlight

   • Use integer ID comparison instead of pointer equality
   • Add 1,000 result limit to prevent excessive iterations
   • Early termination when limit reached

3. Cache invalidation strategy

   • Clear cache on recompilePartially()
   • Clear cache on recompileFullProject()
   • Ensures cache remains accurate after file changes

Performance impact:

• ideUse/ideDus: O(n) -> O(1) for finding all symbol references
• ideHighlight: Faster comparison + early termination
• First lookup builds cache (one-time cost)
• Subsequent lookups are instant hash table access

This significantly improves LSP responsiveness for:

• Find all references (ideUse/ideDus)
• Highlight symbol in file (ideHighlight)
• Large codebases with thousands of symbols

[Araq]

My educated guess: It doesn't improve any query performance in reality and nimsuggest takes up even more memory than it already does.